As space structures grow in scale and modularity, their assembly in orbit poses a considerable challenge for dynamic modelling and control technology. In this context, this paper proposes an optimized multi-robot collaboration strategy for the assembly of large space structures, reducing vibration impact and improving assembly accuracy. In the proposed strategy, a three-branch humanoid robot whose dynamic model is established using the Newton-Euler method is employed for module assembly. A dynamic model of the assembled structures is simplified using the Dynamic Substructure Method wherein the essential low-frequency modes are obtained from the analysis results based on the Finite Element Method (FEM). In the assembly process, before proceeding to the next module, vibrations of the assembled structures are required to be reduced. To reduce these vibrations, a genetic algorithm is used in the optimization of the assembly sequence. Numerical results show that more than 50% of the vibrations can be reduced, and as the number of robots increases, the effect of suppressing vibration becomes more obvious. In a word, this work provides a feasible solution for large space structure assembly, both improving vibration suppression and assembly properties.